4 / 2024-03-13 21:55:24
MSD-YOLO:A Novel Method for Detecting Microscopic Surface Defects in Industrial Products
Defect detection,Computer vision,YOLO,Small object detection,Metal spray painting surface defects
Draft Pending
Yan Zhibo / Zhejiang Normal University
Zhang Teng / Zhejiang Normal University
Zhang Xiaodong / Shaanxi Key Laboratory of Intelligent Robot
Wang Dongyun / Zhejiang Normal University
Metal spray-painted thermal mugs are industrially manufactured products massively produced worldwide. The quality of the surface paint of the metal spray-painted thermal mugs is crucial for enhancing customer satisfaction and reducing product return rates. A novel surface defect target detection network named MSD-YOLO was proposed in this paper. It is based on the YOLO object detection framework, combined with advanced multi-scale feature fusion modules and lightweight convolutional modules, so as to detect microscopic defects while controlling computational costs. Specifically, the sensitivity to microscopic defect features is increased by adding advanced multi-scale feature fusion module. Experiments showed that MSD-YOLO had excellent sensitivity to microscopic defects and real-time performance. Compared with the original YOLO, the index of F1@0.5 and mAP@0.5 of MDSD-YOLO in micro defect target detection increased by 6.8% and 11.6% respectively. It is not only more suitable for detecting microscopic defects, but also easy to deploy in industrial control detection equipment with limited computing power. Furthermore, this study also contributes a dataset of surface defects on metal spray-painted thermal mugs, which enriches the data in the field of surface defects of industrial products.
Important Date
  • Conference Date

    Jun 14

    2024

    to

    Jun 16

    2024

  • Jun 16 2024

    Registration deadline

Sponsored By
IEEE Computational Intelligence Society
IEEE Instrumentation and Measurement Society